Chapter 1. This chapter proposes the Prior Adaptive Bayes classifier (PAB classifier), a new classifier to assign words appearing in a text to their respective topics. It is an adaption of the Bayes classifier where the prior probabilities of topics are replaced with the corresponding posterior probabilities associated with the surrounding words. We carried out experiments on a dataset usually used to test text algorithms, showing that adapting the priors to the corresponding posteriors given that the preceding words occurred allows us to obtain a significant improvement over the original classifier. Moreover, while for the original classifier the accuracy dropped drastically by adding another class, the PAB classifier continued to maintain good performance. Finally, we observed a further improvement in terms of accuracy considering not only the preceding words but also the following words.
From EU Recovery and Resilience Plans to Italian Special Economic Zones: First Systematic Analyses
MUSTICA, Paolo
2024-01-11
Abstract
Chapter 1. This chapter proposes the Prior Adaptive Bayes classifier (PAB classifier), a new classifier to assign words appearing in a text to their respective topics. It is an adaption of the Bayes classifier where the prior probabilities of topics are replaced with the corresponding posterior probabilities associated with the surrounding words. We carried out experiments on a dataset usually used to test text algorithms, showing that adapting the priors to the corresponding posteriors given that the preceding words occurred allows us to obtain a significant improvement over the original classifier. Moreover, while for the original classifier the accuracy dropped drastically by adding another class, the PAB classifier continued to maintain good performance. Finally, we observed a further improvement in terms of accuracy considering not only the preceding words but also the following words.File | Dimensione | Formato | |
---|---|---|---|
Tesi_dottorato_Mustica.pdf
accesso aperto
Tipologia:
Tesi di dottorato
Licenza:
Creative commons
Dimensione
1.86 MB
Formato
Adobe PDF
|
1.86 MB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.